Move idle kse wakeup to outside of regions where locks are held.

This eliminates ping-ponging of locks, where the idle KSE wakes
up only to find the lock it needs is being held.  This gives
little or no gain to M:N mode but greatly speeds up 1:1 mode.

Reviewed & Tested by:	davidxu

16 files changed, 260 insertions, 146 deletions

diff --git a/lib/libkse/thread/thr_cancel.c b/lib/libkse/thread/thr_cancel.c
index 731a9aa..581b719 100644
--- a/lib/libkse/thread/thr_cancel.c
+++ b/lib/libkse/thread/thr_cancel.c
@@ -20,6 +20,7 @@ _pthread_cancel(pthread_t pthread)
 {
 	struct pthread *curthread = _get_curthread();
 	struct pthread *joinee = NULL;
+	struct kse_mailbox *kmbx = NULL;
 	int ret;
 
 	if ((ret = _thr_ref_add(curthread, pthread, /*include dead*/0)) == 0) {
@@ -65,7 +66,7 @@ _pthread_cancel(pthread_t pthread)
 				/* Interrupt and resume: */
 				pthread->interrupted = 1;
 				pthread->cancelflags |= THR_CANCELLING;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				break;
 
 			case PS_JOIN:
@@ -73,7 +74,7 @@ _pthread_cancel(pthread_t pthread)
 				joinee = pthread->join_status.thread;
 				pthread->join_status.thread = NULL;
 				pthread->cancelflags |= THR_CANCELLING;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				if ((joinee != NULL) &&
 				    (pthread->kseg == joinee->kseg)) {
 					/* Remove the joiner from the joinee. */
@@ -97,7 +98,7 @@ _pthread_cancel(pthread_t pthread)
 				 */
 				pthread->interrupted = 1;
 				pthread->cancelflags |= THR_CANCEL_NEEDED;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				pthread->continuation = finish_cancellation;
 				break;
 
@@ -120,6 +121,8 @@ _pthread_cancel(pthread_t pthread)
 		 */
 		THR_SCHED_UNLOCK(curthread, pthread);
 		_thr_ref_delete(curthread, pthread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 
 		if ((joinee != NULL) &&
 		    (_thr_ref_add(curthread, joinee, /* include dead */1) == 0)) {
diff --git a/lib/libkse/thread/thr_cond.c b/lib/libkse/thread/thr_cond.c
index 9703da6..0d6d914 100644
--- a/lib/libkse/thread/thr_cond.c
+++ b/lib/libkse/thread/thr_cond.c
@@ -584,6 +584,7 @@ _pthread_cond_signal(pthread_cond_t * cond)
 {
 	struct pthread	*curthread = _get_curthread();
 	struct pthread	*pthread;
+	struct kse_mailbox *kmbx;
 	int		rval = 0;
 
 	THR_ASSERT(curthread->locklevel == 0,
@@ -619,8 +620,10 @@ _pthread_cond_signal(pthread_cond_t * cond)
 				    (pthread->active_priority >
 				    curthread->active_priority))
 					curthread->critical_yield = 1;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				THR_SCHED_UNLOCK(curthread, pthread);
+				if (kmbx != NULL)
+					kse_wakeup(kmbx);
 			}
 			/* Check for no more waiters: */
 			if (TAILQ_FIRST(&(*cond)->c_queue) == NULL)
@@ -649,6 +652,7 @@ _pthread_cond_broadcast(pthread_cond_t * cond)
 {
 	struct pthread	*curthread = _get_curthread();
 	struct pthread	*pthread;
+	struct kse_mailbox *kmbx;
 	int		rval = 0;
 
 	THR_ASSERT(curthread->locklevel == 0,
@@ -682,8 +686,10 @@ _pthread_cond_broadcast(pthread_cond_t * cond)
 				    (pthread->active_priority >
 				    curthread->active_priority))
 					curthread->critical_yield = 1;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				THR_SCHED_UNLOCK(curthread, pthread);
+				if (kmbx != NULL)
+					kse_wakeup(kmbx);
 			}
 
 			/* There are no more waiting threads: */
diff --git a/lib/libkse/thread/thr_detach.c b/lib/libkse/thread/thr_detach.c
index 0b63dac..0159e53 100644
--- a/lib/libkse/thread/thr_detach.c
+++ b/lib/libkse/thread/thr_detach.c
@@ -43,6 +43,7 @@ int
 _pthread_detach(pthread_t pthread)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx = NULL;
 	struct pthread *joiner;
 	int rval = 0;
 
@@ -83,12 +84,11 @@ _pthread_detach(pthread_t pthread)
 				joiner->join_status.ret = NULL;
 				joiner->join_status.thread = NULL;
 
-				_thr_setrunnable_unlocked(joiner);
+				kmbx = _thr_setrunnable_unlocked(joiner);
 			}
 			joiner = NULL;
 		}
 		THR_SCHED_UNLOCK(curthread, pthread);
-
 		/* See if there is a thread waiting in pthread_join(): */
 		if ((joiner != NULL) &&
 		    (_thr_ref_add(curthread, joiner, 0) == 0)) {
@@ -102,12 +102,14 @@ _pthread_detach(pthread_t pthread)
 				joiner->join_status.ret = NULL;
 				joiner->join_status.thread = NULL;
 
-				_thr_setrunnable_unlocked(joiner);
+				kmbx = _thr_setrunnable_unlocked(joiner);
 			}
 			THR_SCHED_UNLOCK(curthread, joiner);
 			_thr_ref_delete(curthread, joiner);
 		}
 		_thr_ref_delete(curthread, pthread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 	}
 
 	/* Return the completion status: */
diff --git a/lib/libkse/thread/thr_kern.c b/lib/libkse/thread/thr_kern.c
index 51aab62..a08c677 100644
--- a/lib/libkse/thread/thr_kern.c
+++ b/lib/libkse/thread/thr_kern.c
@@ -142,7 +142,7 @@ static void	kseg_init(struct kse_group *kseg);
 static void	kseg_reinit(struct kse_group *kseg);
 static void	kse_waitq_insert(struct pthread *thread);
 static void	kse_wakeup_multi(struct kse *curkse);
-static void	kse_wakeup_one(struct pthread *thread);
+static struct kse_mailbox *kse_wakeup_one(struct pthread *thread);
 static void	thr_cleanup(struct kse *kse, struct pthread *curthread);
 static void	thr_link(struct pthread *thread);
 static void	thr_resume_wrapper(int sig, siginfo_t *, ucontext_t *);
@@ -341,7 +341,7 @@ _kse_single_thread(struct pthread *curthread)
 #else
 	if (__isthreaded)
 		_thr_signal_deinit();
-	_ksd_readandclear_tmbx();
+	_ksd_set_tmbx(NULL);
 	__isthreaded   = 0;
 	active_threads = 0;
 #endif
@@ -505,10 +505,9 @@ _thr_lock_wait(struct lock *lock, struct lockuser *lu)
 	struct pthread *curthread = (struct pthread *)lu->lu_private;
 
 	do {
-		THR_SCHED_LOCK(curthread, curthread);
+		THR_LOCK_SWITCH(curthread);
 		THR_SET_STATE(curthread, PS_LOCKWAIT);
-		THR_SCHED_UNLOCK(curthread, curthread);
-		_thr_sched_switch(curthread);
+		_thr_sched_switch_unlocked(curthread);
 	} while (!_LCK_GRANTED(lu));
 }
 
@@ -517,14 +516,17 @@ _thr_lock_wakeup(struct lock *lock, struct lockuser *lu)
 {
 	struct pthread *thread;
 	struct pthread *curthread;
+	struct kse_mailbox *kmbx;
 
 	curthread = _get_curthread();
 	thread = (struct pthread *)_LCK_GET_PRIVATE(lu);
 
 	THR_SCHED_LOCK(curthread, thread);
 	_lock_grant(lock, lu);
-	_thr_setrunnable_unlocked(thread);
+	kmbx = _thr_setrunnable_unlocked(thread);
 	THR_SCHED_UNLOCK(curthread, thread);
+	if (kmbx != NULL)
+		kse_wakeup(kmbx);
 }
 
 kse_critical_t
@@ -532,7 +534,8 @@ _kse_critical_enter(void)
 {
 	kse_critical_t crit;
 
-	crit = _ksd_readandclear_tmbx();
+	crit = _ksd_get_tmbx();
+	_ksd_set_tmbx(NULL);
 	return (crit);
 }
 
@@ -841,8 +844,8 @@ kse_sched_single(struct kse *curkse)
 				if (SIGISMEMBER(curthread->sigmask, i))
 					continue;
 				if (SIGISMEMBER(curthread->sigpend, i))
-					_thr_sig_add(curthread, i, 
-						&curthread->siginfo[i-1]);
+					(void)_thr_sig_add(curthread, i, 
+					    &curthread->siginfo[i-1]);
 			}
 			__sys_sigprocmask(SIG_SETMASK, &curthread->sigmask,
 				NULL);
@@ -1142,6 +1145,7 @@ static void
 thr_cleanup(struct kse *curkse, struct pthread *thread)
 {
 	struct pthread *joiner;
+	struct kse_mailbox *kmbx = NULL;
 	int sys_scope;
 
 	if ((joiner = thread->joiner) != NULL) {
@@ -1150,7 +1154,7 @@ thr_cleanup(struct kse *curkse, struct pthread *thread)
 			if (joiner->join_status.thread == thread) {
 				joiner->join_status.thread = NULL;
 				joiner->join_status.ret = thread->ret;
-				_thr_setrunnable_unlocked(joiner);
+				(void)_thr_setrunnable_unlocked(joiner);
 			}
 		} else {
 			KSE_SCHED_UNLOCK(curkse, curkse->k_kseg);
@@ -1160,10 +1164,12 @@ thr_cleanup(struct kse *curkse, struct pthread *thread)
 				if (joiner->join_status.thread == thread) {
 					joiner->join_status.thread = NULL;
 					joiner->join_status.ret = thread->ret;
-					_thr_setrunnable_unlocked(joiner);
+					kmbx = _thr_setrunnable_unlocked(joiner);
 				}
 				KSE_SCHED_UNLOCK(curkse, joiner->kseg);
 				_thr_ref_delete(thread, joiner);
+				if (kmbx != NULL)
+					kse_wakeup(kmbx);
 			}
 			KSE_SCHED_LOCK(curkse, curkse->k_kseg);
 		}
@@ -1436,7 +1442,8 @@ kse_check_completed(struct kse *kse)
 				if (SIGISMEMBER(thread->sigmask, sig))
 					SIGADDSET(thread->sigpend, sig);
 				else
-					_thr_sig_add(thread, sig, &thread->tmbx.tm_syncsig);
+					(void)_thr_sig_add(thread, sig,
+					    &thread->tmbx.tm_syncsig);
 				thread->tmbx.tm_syncsig.si_signo = 0;
 			}
 			completed = completed->tm_next;
@@ -1653,10 +1660,11 @@ kse_switchout_thread(struct kse *kse, struct pthread *thread)
 			if (SIGISMEMBER(thread->sigmask, i))
 				continue;
 			if (SIGISMEMBER(thread->sigpend, i))
-				_thr_sig_add(thread, i, &thread->siginfo[i-1]);
+				(void)_thr_sig_add(thread, i,
+				    &thread->siginfo[i-1]);
 			else if (SIGISMEMBER(_thr_proc_sigpending, i) &&
 				_thr_getprocsig_unlocked(i, &siginfo)) {
-				_thr_sig_add(thread, i, &siginfo);
+				(void)_thr_sig_add(thread, i, &siginfo);
 			}
 		}
 		KSE_LOCK_RELEASE(kse, &_thread_signal_lock);
@@ -1823,23 +1831,31 @@ void
 _thr_setrunnable(struct pthread *curthread, struct pthread *thread)
 {
 	kse_critical_t crit;
+	struct kse_mailbox *kmbx;
 
 	crit = _kse_critical_enter();
 	KSE_SCHED_LOCK(curthread->kse, thread->kseg);
-	_thr_setrunnable_unlocked(thread);
+	kmbx = _thr_setrunnable_unlocked(thread);
 	KSE_SCHED_UNLOCK(curthread->kse, thread->kseg);
 	_kse_critical_leave(crit);
+	if (kmbx != NULL)
+		kse_wakeup(kmbx);
 }
 
-void
+struct kse_mailbox *
 _thr_setrunnable_unlocked(struct pthread *thread)
 {
+	struct kse_mailbox *kmbx = NULL;
+
 	if ((thread->kseg->kg_flags & KGF_SINGLE_THREAD) != 0) {
 		/* No silly queues for these threads. */
 		if ((thread->flags & THR_FLAGS_SUSPENDED) != 0)
 			THR_SET_STATE(thread, PS_SUSPENDED);
-		else
+		else {
 			THR_SET_STATE(thread, PS_RUNNING);
+			kmbx = kse_wakeup_one(thread);
+		}
+
 	} else if (thread->state != PS_RUNNING) {
 		if ((thread->flags & THR_FLAGS_IN_WAITQ) != 0)
 			KSE_WAITQ_REMOVE(thread->kse, thread);
@@ -1850,25 +1866,31 @@ _thr_setrunnable_unlocked(struct pthread *thread)
 			if ((thread->blocked == 0) && (thread->active == 0) &&
 			    (thread->flags & THR_FLAGS_IN_RUNQ) == 0)
 				THR_RUNQ_INSERT_TAIL(thread);
+			/*
+			 * XXX - Threads are not yet assigned to specific
+			 *       KSEs; they are assigned to the KSEG.  So
+			 *       the fact that a thread's KSE is waiting
+			 *       doesn't necessarily mean that it will be
+			 *       the KSE that runs the thread after the
+			 *       lock is granted.  But we don't know if the
+			 *       other KSEs within the same KSEG are also
+			 *       in a waiting state or not so we err on the
+			 *       side of caution and wakeup the thread's
+			 *       last known KSE.  We ensure that the
+			 *       threads KSE doesn't change while it's
+			 *       scheduling lock is held so it is safe to
+			 *       reference it (the KSE).  If the KSE wakes
+			 *       up and doesn't find any more work it will
+			 *       again go back to waiting so no harm is
+			 *       done.
+			 */
+			kmbx = kse_wakeup_one(thread);
 		}
 	}
-        /*
-         * XXX - Threads are not yet assigned to specific KSEs; they are
-         *       assigned to the KSEG.  So the fact that a thread's KSE is
-         *       waiting doesn't necessarily mean that it will be the KSE
-         *       that runs the thread after the lock is granted.  But we
-         *       don't know if the other KSEs within the same KSEG are
-         *       also in a waiting state or not so we err on the side of
-         *       caution and wakeup the thread's last known KSE.  We
-         *       ensure that the threads KSE doesn't change while it's
-         *       scheduling lock is held so it is safe to reference it
-         *       (the KSE).  If the KSE wakes up and doesn't find any more
-         *       work it will again go back to waiting so no harm is done.
-         */
-	kse_wakeup_one(thread);
+	return (kmbx);
 }
 
-static void
+static struct kse_mailbox *
 kse_wakeup_one(struct pthread *thread)
 {
 	struct kse *ke;
@@ -1876,17 +1898,17 @@ kse_wakeup_one(struct pthread *thread)
 	if (KSE_IS_IDLE(thread->kse)) {
 		KSE_CLEAR_IDLE(thread->kse);
 		thread->kseg->kg_idle_kses--;
-		KSE_WAKEUP(thread->kse);
+		return (&thread->kse->k_mbx);
 	} else {
 		TAILQ_FOREACH(ke, &thread->kseg->kg_kseq, k_kgqe) {
 			if (KSE_IS_IDLE(ke)) {
 				KSE_CLEAR_IDLE(ke);
 				ke->k_kseg->kg_idle_kses--;
-				KSE_WAKEUP(ke);
-				return;
+				return (&ke->k_mbx);
 			}
 		}
 	}
+	return (NULL);
 }
 
 static void
diff --git a/lib/libkse/thread/thr_mutex.c b/lib/libkse/thread/thr_mutex.c
index c2c28e0..7e8de44 100644
--- a/lib/libkse/thread/thr_mutex.c
+++ b/lib/libkse/thread/thr_mutex.c
@@ -67,7 +67,8 @@
 /*
  * Prototypes
  */
-static void		mutex_handoff(struct pthread *, struct pthread_mutex *);
+static struct kse_mailbox *mutex_handoff(struct pthread *,
+			    struct pthread_mutex *);
 static inline int	mutex_self_trylock(struct pthread *, pthread_mutex_t);
 static inline int	mutex_self_lock(struct pthread *, pthread_mutex_t);
 static int		mutex_unlock_common(pthread_mutex_t *, int);
@@ -860,6 +861,7 @@ static int
 mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx = NULL;
 	int ret = 0;
 
 	if (m == NULL || *m == NULL)
@@ -904,7 +906,7 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 				 * Hand off the mutex to the next waiting
 				 * thread:
 				 */
-				mutex_handoff(curthread, *m);
+				kmbx = mutex_handoff(curthread, *m);
 			}
 			break;
 
@@ -961,7 +963,7 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 				 * Hand off the mutex to the next waiting
 				 * thread:
 				 */
-				mutex_handoff(curthread, *m);
+				kmbx = mutex_handoff(curthread, *m);
 			}
 			break;
 
@@ -1017,7 +1019,7 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 				 * Hand off the mutex to the next waiting
 				 * thread:
 				 */
-				mutex_handoff(curthread, *m);
+				kmbx = mutex_handoff(curthread, *m);
 			}
 			break;
 
@@ -1034,6 +1036,8 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 
 		/* Unlock the mutex structure: */
 		THR_LOCK_RELEASE(curthread, &(*m)->m_lock);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 	}
 
 	/* Return the completion status: */
@@ -1460,9 +1464,10 @@ _mutex_lock_backout(struct pthread *curthread)
  * is necessary to lock the thread's scheduling queue while also
  * holding the mutex lock.
  */
-static void
+static struct kse_mailbox *
 mutex_handoff(struct pthread *curthread, struct pthread_mutex *mutex)
 {
+	struct kse_mailbox *kmbx = NULL;
 	struct pthread *pthread;
 
 	/* Keep dequeueing until we find a valid thread: */
@@ -1564,7 +1569,7 @@ mutex_handoff(struct pthread *curthread, struct pthread_mutex *mutex)
 		}
 
 		/* Make the thread runnable and unlock the scheduling queue: */
-		_thr_setrunnable_unlocked(pthread);
+		kmbx = _thr_setrunnable_unlocked(pthread);
 
 		/* Add a preemption point. */
 		if ((curthread->kseg == pthread->kseg) &&
@@ -1583,6 +1588,7 @@ mutex_handoff(struct pthread *curthread, struct pthread_mutex *mutex)
 	if ((pthread == NULL) && (mutex->m_protocol == PTHREAD_PRIO_INHERIT))
 		/* This mutex has no priority: */
 		mutex->m_prio = 0;
+	return (kmbx);
 }
 
 /*
diff --git a/lib/libkse/thread/thr_private.h b/lib/libkse/thread/thr_private.h
index 21026b7..90dbed7 100644
--- a/lib/libkse/thread/thr_private.h
+++ b/lib/libkse/thread/thr_private.h
@@ -1106,8 +1106,8 @@ void	_thr_ref_delete(struct pthread *, struct pthread *);
 int	_thr_schedule_add(struct pthread *, struct pthread *);
 void	_thr_schedule_remove(struct pthread *, struct pthread *);
 void	_thr_setrunnable(struct pthread *curthread, struct pthread *thread);
-void	_thr_setrunnable_unlocked(struct pthread *thread);
-void	_thr_sig_add(struct pthread *, int, siginfo_t *);
+struct kse_mailbox *_thr_setrunnable_unlocked(struct pthread *thread);
+struct kse_mailbox *_thr_sig_add(struct pthread *, int, siginfo_t *);
 void	_thr_sig_dispatch(struct kse *, int, siginfo_t *);
 int	_thr_stack_alloc(struct pthread_attr *);
 void	_thr_stack_free(struct pthread_attr *);
diff --git a/lib/libkse/thread/thr_resume_np.c b/lib/libkse/thread/thr_resume_np.c
index 1276311..ba9b911 100644
--- a/lib/libkse/thread/thr_resume_np.c
+++ b/lib/libkse/thread/thr_resume_np.c
@@ -35,7 +35,7 @@
 #include <pthread.h>
 #include "thr_private.h"
 
-static void resume_common(struct pthread *);
+static struct kse_mailbox *resume_common(struct pthread *);
 
 __weak_reference(_pthread_resume_np, pthread_resume_np);
 __weak_reference(_pthread_resume_all_np, pthread_resume_all_np);
@@ -46,15 +46,18 @@ int
 _pthread_resume_np(pthread_t thread)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx;
 	int ret;
 
 	/* Add a reference to the thread: */
 	if ((ret = _thr_ref_add(curthread, thread, /*include dead*/0)) == 0) {
 		/* Lock the threads scheduling queue: */
 		THR_SCHED_LOCK(curthread, thread);
-		resume_common(thread);
+		kmbx = resume_common(thread);
 		THR_SCHED_UNLOCK(curthread, thread);
 		_thr_ref_delete(curthread, thread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 	}
 	return (ret);
 }
@@ -64,6 +67,7 @@ _pthread_resume_all_np(void)
 {
 	struct pthread *curthread = _get_curthread();
 	struct pthread *thread;
+	struct kse_mailbox *kmbx;
 	kse_critical_t crit;
 
 	/* Take the thread list lock: */
@@ -73,8 +77,10 @@ _pthread_resume_all_np(void)
 	TAILQ_FOREACH(thread, &_thread_list, tle) {
 		if (thread != curthread) {
 			THR_SCHED_LOCK(curthread, thread);
-			resume_common(thread);
+			kmbx = resume_common(thread);
 			THR_SCHED_UNLOCK(curthread, thread);
+			if (kmbx != NULL)
+				kse_wakeup(kmbx);
 		}
 	}
 
@@ -83,7 +89,7 @@ _pthread_resume_all_np(void)
 	_kse_critical_leave(crit);
 }
 
-static void
+static struct kse_mailbox *
 resume_common(struct pthread *thread)
 {
 	/* Clear the suspend flag: */
@@ -95,5 +101,7 @@ resume_common(struct pthread *thread)
 	 * state to running and insert it into the run queue.
 	 */
 	if (thread->state == PS_SUSPENDED)
-		_thr_setrunnable_unlocked(thread);
+		return (_thr_setrunnable_unlocked(thread));
+	else
+		return (NULL);
 }
diff --git a/lib/libkse/thread/thr_sig.c b/lib/libkse/thread/thr_sig.c
index 1211ba3..fc075af 100644
--- a/lib/libkse/thread/thr_sig.c
+++ b/lib/libkse/thread/thr_sig.c
@@ -249,6 +249,7 @@ _thr_start_sig_daemon(void)
 void
 _thr_sig_dispatch(struct kse *curkse, int sig, siginfo_t *info)
 {
+	struct kse_mailbox *kmbx;
 	struct pthread *thread;
 
 	DBG_MSG(">>> _thr_sig_dispatch(%d)\n", sig);
@@ -280,9 +281,11 @@ _thr_sig_dispatch(struct kse *curkse, int sig, siginfo_t *info)
 			KSE_SCHED_UNLOCK(curkse, thread->kseg);
 			_thr_ref_delete(NULL, thread);
 		} else {
-			_thr_sig_add(thread, sig, info);
+			kmbx = _thr_sig_add(thread, sig, info);
 			KSE_SCHED_UNLOCK(curkse, thread->kseg);
 			_thr_ref_delete(NULL, thread);
+			if (kmbx != NULL)
+				kse_wakeup(kmbx);
 			break;
 		}
 	}
@@ -533,6 +536,7 @@ _thr_getprocsig_unlocked(int sig, siginfo_t *siginfo)
 struct pthread *
 thr_sig_find(struct kse *curkse, int sig, siginfo_t *info)
 {
+	struct kse_mailbox *kmbx = NULL;
 	struct pthread	*pthread;
 	struct pthread	*suspended_thread, *signaled_thread;
 	__siginfohandler_t *sigfunc;
@@ -582,7 +586,7 @@ thr_sig_find(struct kse *curkse, int sig, siginfo_t *info)
 				/*  where to put siginfo ? */
 				*(pthread->data.sigwaitinfo) = si;
 				pthread->sigmask = pthread->oldsigmask;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 			}
 			KSE_SCHED_UNLOCK(curkse, pthread->kseg);
 			/*
@@ -596,6 +600,8 @@ thr_sig_find(struct kse *curkse, int sig, siginfo_t *info)
 			 * to the process pending set.
 			 */
 			KSE_LOCK_RELEASE(curkse, &_thread_list_lock);
+			if (kmbx != NULL)
+				kse_wakeup(kmbx);
 			return (NULL);
 		} else if (!SIGISMEMBER(pthread->sigmask, sig) ||
 			   (!SIGISMEMBER(pthread->oldsigmask, sig) &&
@@ -811,15 +817,16 @@ handle_special_signals(struct kse *curkse, int sig)
  *
  * This must be called with the thread's scheduling lock held.
  */
-void
+struct kse_mailbox *
 _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 {
+	siginfo_t siginfo;
+	struct kse *curkse;
+	struct kse_mailbox *kmbx = NULL;
+	struct pthread *curthread = _get_curthread();
 	int	restart;
 	int	suppress_handler = 0;
 	int	fromproc = 0;
-	struct  pthread *curthread = _get_curthread();
-	struct	kse *curkse;
-	siginfo_t siginfo;
 
 	DBG_MSG(">>> _thr_sig_add %p (%d)\n", pthread, sig);
 
@@ -829,12 +836,12 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 
 	if (pthread->state == PS_DEAD || pthread->state == PS_DEADLOCK ||
 	    pthread->state == PS_STATE_MAX)
-	    	return; /* return false */
+	    	return (NULL); /* return false */
 
 	if ((pthread->attr.flags & PTHREAD_SCOPE_SYSTEM) &&
 	    (curthread != pthread)) {
 	    	PANIC("Please use _thr_send_sig for bound thread");
-		return;
+		return (NULL);
 	}
 
 	if (pthread->curframe == NULL ||
@@ -851,7 +858,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 					 sizeof(*info));
 		} else {
 			if (!_thr_getprocsig(sig, &pthread->siginfo[sig-1]))
-				return;
+				return (NULL);
 			SIGADDSET(pthread->sigpend, sig);
 		}
 		if (!SIGISMEMBER(pthread->sigmask, sig)) {
@@ -867,7 +874,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 		/* if process signal not exists, just return */
 		if (fromproc) {
 			if (!_thr_getprocsig(sig, &siginfo))
-				return;
+				return (NULL);
 			info = &siginfo;
 		}
 		/*
@@ -877,7 +884,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 		case PS_DEAD:
 		case PS_DEADLOCK:
 		case PS_STATE_MAX:
-			return;	/* XXX return false */
+			return (NULL);	/* XXX return false */
 		case PS_LOCKWAIT:
 		case PS_SUSPENDED:
 			/*
@@ -937,7 +944,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 				*(pthread->data.sigwaitinfo) = pthread->siginfo[sig-1];
 				pthread->sigmask = pthread->oldsigmask;
 				/* Make the thread runnable: */
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 			} else {
 				/* Increment the pending signal count. */
 				SIGADDSET(pthread->sigpend, sig);
@@ -945,11 +952,10 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 					pthread->check_pending = 1;
 					pthread->interrupted = 1;
 					pthread->sigmask = pthread->oldsigmask;
-					_thr_setrunnable_unlocked(pthread);
+					kmbx = _thr_setrunnable_unlocked(pthread);
 				}
 			}
-		
-			return;
+			return (kmbx);
 		}
 
 		SIGADDSET(pthread->sigpend, sig);
@@ -967,11 +973,12 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 			if (pthread->flags & THR_FLAGS_IN_RUNQ)
 				THR_RUNQ_REMOVE(pthread);
 			pthread->active_priority |= THR_SIGNAL_PRIORITY;
-			_thr_setrunnable_unlocked(pthread);
+			kmbx = _thr_setrunnable_unlocked(pthread);
 		} else {
 			pthread->check_pending = 1;
 		}
 	}
+	return (kmbx);
 }
 
 /*
@@ -981,6 +988,7 @@ void
 _thr_sig_send(struct pthread *pthread, int sig)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx;
 
 	if (pthread->attr.flags & PTHREAD_SCOPE_SYSTEM) {
 		kse_thr_interrupt(&pthread->tmbx, KSE_INTR_SENDSIG, sig);
@@ -990,8 +998,10 @@ _thr_sig_send(struct pthread *pthread, int sig)
 	/* Lock the scheduling queue of the target thread. */
 	THR_SCHED_LOCK(curthread, pthread);
 	if (_thread_sigact[sig - 1].sa_handler != SIG_IGN) {
-		_thr_sig_add(pthread, sig, NULL);
+		kmbx = _thr_sig_add(pthread, sig, NULL);
 		THR_SCHED_UNLOCK(curthread, pthread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 		/* XXX
 		 * If thread sent signal to itself, check signals now.
 		 * It is not really needed, _kse_critical_leave should
diff --git a/lib/libpthread/thread/thr_cancel.c b/lib/libpthread/thread/thr_cancel.c
index 731a9aa..581b719 100644
--- a/lib/libpthread/thread/thr_cancel.c
+++ b/lib/libpthread/thread/thr_cancel.c
@@ -20,6 +20,7 @@ _pthread_cancel(pthread_t pthread)
 {
 	struct pthread *curthread = _get_curthread();
 	struct pthread *joinee = NULL;
+	struct kse_mailbox *kmbx = NULL;
 	int ret;
 
 	if ((ret = _thr_ref_add(curthread, pthread, /*include dead*/0)) == 0) {
@@ -65,7 +66,7 @@ _pthread_cancel(pthread_t pthread)
 				/* Interrupt and resume: */
 				pthread->interrupted = 1;
 				pthread->cancelflags |= THR_CANCELLING;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				break;
 
 			case PS_JOIN:
@@ -73,7 +74,7 @@ _pthread_cancel(pthread_t pthread)
 				joinee = pthread->join_status.thread;
 				pthread->join_status.thread = NULL;
 				pthread->cancelflags |= THR_CANCELLING;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				if ((joinee != NULL) &&
 				    (pthread->kseg == joinee->kseg)) {
 					/* Remove the joiner from the joinee. */
@@ -97,7 +98,7 @@ _pthread_cancel(pthread_t pthread)
 				 */
 				pthread->interrupted = 1;
 				pthread->cancelflags |= THR_CANCEL_NEEDED;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				pthread->continuation = finish_cancellation;
 				break;
 
@@ -120,6 +121,8 @@ _pthread_cancel(pthread_t pthread)
 		 */
 		THR_SCHED_UNLOCK(curthread, pthread);
 		_thr_ref_delete(curthread, pthread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 
 		if ((joinee != NULL) &&
 		    (_thr_ref_add(curthread, joinee, /* include dead */1) == 0)) {
diff --git a/lib/libpthread/thread/thr_cond.c b/lib/libpthread/thread/thr_cond.c
index 9703da6..0d6d914 100644
--- a/lib/libpthread/thread/thr_cond.c
+++ b/lib/libpthread/thread/thr_cond.c
@@ -584,6 +584,7 @@ _pthread_cond_signal(pthread_cond_t * cond)
 {
 	struct pthread	*curthread = _get_curthread();
 	struct pthread	*pthread;
+	struct kse_mailbox *kmbx;
 	int		rval = 0;
 
 	THR_ASSERT(curthread->locklevel == 0,
@@ -619,8 +620,10 @@ _pthread_cond_signal(pthread_cond_t * cond)
 				    (pthread->active_priority >
 				    curthread->active_priority))
 					curthread->critical_yield = 1;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				THR_SCHED_UNLOCK(curthread, pthread);
+				if (kmbx != NULL)
+					kse_wakeup(kmbx);
 			}
 			/* Check for no more waiters: */
 			if (TAILQ_FIRST(&(*cond)->c_queue) == NULL)
@@ -649,6 +652,7 @@ _pthread_cond_broadcast(pthread_cond_t * cond)
 {
 	struct pthread	*curthread = _get_curthread();
 	struct pthread	*pthread;
+	struct kse_mailbox *kmbx;
 	int		rval = 0;
 
 	THR_ASSERT(curthread->locklevel == 0,
@@ -682,8 +686,10 @@ _pthread_cond_broadcast(pthread_cond_t * cond)
 				    (pthread->active_priority >
 				    curthread->active_priority))
 					curthread->critical_yield = 1;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 				THR_SCHED_UNLOCK(curthread, pthread);
+				if (kmbx != NULL)
+					kse_wakeup(kmbx);
 			}
 
 			/* There are no more waiting threads: */
diff --git a/lib/libpthread/thread/thr_detach.c b/lib/libpthread/thread/thr_detach.c
index 0b63dac..0159e53 100644
--- a/lib/libpthread/thread/thr_detach.c
+++ b/lib/libpthread/thread/thr_detach.c
@@ -43,6 +43,7 @@ int
 _pthread_detach(pthread_t pthread)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx = NULL;
 	struct pthread *joiner;
 	int rval = 0;
 
@@ -83,12 +84,11 @@ _pthread_detach(pthread_t pthread)
 				joiner->join_status.ret = NULL;
 				joiner->join_status.thread = NULL;
 
-				_thr_setrunnable_unlocked(joiner);
+				kmbx = _thr_setrunnable_unlocked(joiner);
 			}
 			joiner = NULL;
 		}
 		THR_SCHED_UNLOCK(curthread, pthread);
-
 		/* See if there is a thread waiting in pthread_join(): */
 		if ((joiner != NULL) &&
 		    (_thr_ref_add(curthread, joiner, 0) == 0)) {
@@ -102,12 +102,14 @@ _pthread_detach(pthread_t pthread)
 				joiner->join_status.ret = NULL;
 				joiner->join_status.thread = NULL;
 
-				_thr_setrunnable_unlocked(joiner);
+				kmbx = _thr_setrunnable_unlocked(joiner);
 			}
 			THR_SCHED_UNLOCK(curthread, joiner);
 			_thr_ref_delete(curthread, joiner);
 		}
 		_thr_ref_delete(curthread, pthread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 	}
 
 	/* Return the completion status: */
diff --git a/lib/libpthread/thread/thr_kern.c b/lib/libpthread/thread/thr_kern.c
index 51aab62..a08c677 100644
--- a/lib/libpthread/thread/thr_kern.c
+++ b/lib/libpthread/thread/thr_kern.c
@@ -142,7 +142,7 @@ static void	kseg_init(struct kse_group *kseg);
 static void	kseg_reinit(struct kse_group *kseg);
 static void	kse_waitq_insert(struct pthread *thread);
 static void	kse_wakeup_multi(struct kse *curkse);
-static void	kse_wakeup_one(struct pthread *thread);
+static struct kse_mailbox *kse_wakeup_one(struct pthread *thread);
 static void	thr_cleanup(struct kse *kse, struct pthread *curthread);
 static void	thr_link(struct pthread *thread);
 static void	thr_resume_wrapper(int sig, siginfo_t *, ucontext_t *);
@@ -341,7 +341,7 @@ _kse_single_thread(struct pthread *curthread)
 #else
 	if (__isthreaded)
 		_thr_signal_deinit();
-	_ksd_readandclear_tmbx();
+	_ksd_set_tmbx(NULL);
 	__isthreaded   = 0;
 	active_threads = 0;
 #endif
@@ -505,10 +505,9 @@ _thr_lock_wait(struct lock *lock, struct lockuser *lu)
 	struct pthread *curthread = (struct pthread *)lu->lu_private;
 
 	do {
-		THR_SCHED_LOCK(curthread, curthread);
+		THR_LOCK_SWITCH(curthread);
 		THR_SET_STATE(curthread, PS_LOCKWAIT);
-		THR_SCHED_UNLOCK(curthread, curthread);
-		_thr_sched_switch(curthread);
+		_thr_sched_switch_unlocked(curthread);
 	} while (!_LCK_GRANTED(lu));
 }
 
@@ -517,14 +516,17 @@ _thr_lock_wakeup(struct lock *lock, struct lockuser *lu)
 {
 	struct pthread *thread;
 	struct pthread *curthread;
+	struct kse_mailbox *kmbx;
 
 	curthread = _get_curthread();
 	thread = (struct pthread *)_LCK_GET_PRIVATE(lu);
 
 	THR_SCHED_LOCK(curthread, thread);
 	_lock_grant(lock, lu);
-	_thr_setrunnable_unlocked(thread);
+	kmbx = _thr_setrunnable_unlocked(thread);
 	THR_SCHED_UNLOCK(curthread, thread);
+	if (kmbx != NULL)
+		kse_wakeup(kmbx);
 }
 
 kse_critical_t
@@ -532,7 +534,8 @@ _kse_critical_enter(void)
 {
 	kse_critical_t crit;
 
-	crit = _ksd_readandclear_tmbx();
+	crit = _ksd_get_tmbx();
+	_ksd_set_tmbx(NULL);
 	return (crit);
 }
 
@@ -841,8 +844,8 @@ kse_sched_single(struct kse *curkse)
 				if (SIGISMEMBER(curthread->sigmask, i))
 					continue;
 				if (SIGISMEMBER(curthread->sigpend, i))
-					_thr_sig_add(curthread, i, 
-						&curthread->siginfo[i-1]);
+					(void)_thr_sig_add(curthread, i, 
+					    &curthread->siginfo[i-1]);
 			}
 			__sys_sigprocmask(SIG_SETMASK, &curthread->sigmask,
 				NULL);
@@ -1142,6 +1145,7 @@ static void
 thr_cleanup(struct kse *curkse, struct pthread *thread)
 {
 	struct pthread *joiner;
+	struct kse_mailbox *kmbx = NULL;
 	int sys_scope;
 
 	if ((joiner = thread->joiner) != NULL) {
@@ -1150,7 +1154,7 @@ thr_cleanup(struct kse *curkse, struct pthread *thread)
 			if (joiner->join_status.thread == thread) {
 				joiner->join_status.thread = NULL;
 				joiner->join_status.ret = thread->ret;
-				_thr_setrunnable_unlocked(joiner);
+				(void)_thr_setrunnable_unlocked(joiner);
 			}
 		} else {
 			KSE_SCHED_UNLOCK(curkse, curkse->k_kseg);
@@ -1160,10 +1164,12 @@ thr_cleanup(struct kse *curkse, struct pthread *thread)
 				if (joiner->join_status.thread == thread) {
 					joiner->join_status.thread = NULL;
 					joiner->join_status.ret = thread->ret;
-					_thr_setrunnable_unlocked(joiner);
+					kmbx = _thr_setrunnable_unlocked(joiner);
 				}
 				KSE_SCHED_UNLOCK(curkse, joiner->kseg);
 				_thr_ref_delete(thread, joiner);
+				if (kmbx != NULL)
+					kse_wakeup(kmbx);
 			}
 			KSE_SCHED_LOCK(curkse, curkse->k_kseg);
 		}
@@ -1436,7 +1442,8 @@ kse_check_completed(struct kse *kse)
 				if (SIGISMEMBER(thread->sigmask, sig))
 					SIGADDSET(thread->sigpend, sig);
 				else
-					_thr_sig_add(thread, sig, &thread->tmbx.tm_syncsig);
+					(void)_thr_sig_add(thread, sig,
+					    &thread->tmbx.tm_syncsig);
 				thread->tmbx.tm_syncsig.si_signo = 0;
 			}
 			completed = completed->tm_next;
@@ -1653,10 +1660,11 @@ kse_switchout_thread(struct kse *kse, struct pthread *thread)
 			if (SIGISMEMBER(thread->sigmask, i))
 				continue;
 			if (SIGISMEMBER(thread->sigpend, i))
-				_thr_sig_add(thread, i, &thread->siginfo[i-1]);
+				(void)_thr_sig_add(thread, i,
+				    &thread->siginfo[i-1]);
 			else if (SIGISMEMBER(_thr_proc_sigpending, i) &&
 				_thr_getprocsig_unlocked(i, &siginfo)) {
-				_thr_sig_add(thread, i, &siginfo);
+				(void)_thr_sig_add(thread, i, &siginfo);
 			}
 		}
 		KSE_LOCK_RELEASE(kse, &_thread_signal_lock);
@@ -1823,23 +1831,31 @@ void
 _thr_setrunnable(struct pthread *curthread, struct pthread *thread)
 {
 	kse_critical_t crit;
+	struct kse_mailbox *kmbx;
 
 	crit = _kse_critical_enter();
 	KSE_SCHED_LOCK(curthread->kse, thread->kseg);
-	_thr_setrunnable_unlocked(thread);
+	kmbx = _thr_setrunnable_unlocked(thread);
 	KSE_SCHED_UNLOCK(curthread->kse, thread->kseg);
 	_kse_critical_leave(crit);
+	if (kmbx != NULL)
+		kse_wakeup(kmbx);
 }
 
-void
+struct kse_mailbox *
 _thr_setrunnable_unlocked(struct pthread *thread)
 {
+	struct kse_mailbox *kmbx = NULL;
+
 	if ((thread->kseg->kg_flags & KGF_SINGLE_THREAD) != 0) {
 		/* No silly queues for these threads. */
 		if ((thread->flags & THR_FLAGS_SUSPENDED) != 0)
 			THR_SET_STATE(thread, PS_SUSPENDED);
-		else
+		else {
 			THR_SET_STATE(thread, PS_RUNNING);
+			kmbx = kse_wakeup_one(thread);
+		}
+
 	} else if (thread->state != PS_RUNNING) {
 		if ((thread->flags & THR_FLAGS_IN_WAITQ) != 0)
 			KSE_WAITQ_REMOVE(thread->kse, thread);
@@ -1850,25 +1866,31 @@ _thr_setrunnable_unlocked(struct pthread *thread)
 			if ((thread->blocked == 0) && (thread->active == 0) &&
 			    (thread->flags & THR_FLAGS_IN_RUNQ) == 0)
 				THR_RUNQ_INSERT_TAIL(thread);
+			/*
+			 * XXX - Threads are not yet assigned to specific
+			 *       KSEs; they are assigned to the KSEG.  So
+			 *       the fact that a thread's KSE is waiting
+			 *       doesn't necessarily mean that it will be
+			 *       the KSE that runs the thread after the
+			 *       lock is granted.  But we don't know if the
+			 *       other KSEs within the same KSEG are also
+			 *       in a waiting state or not so we err on the
+			 *       side of caution and wakeup the thread's
+			 *       last known KSE.  We ensure that the
+			 *       threads KSE doesn't change while it's
+			 *       scheduling lock is held so it is safe to
+			 *       reference it (the KSE).  If the KSE wakes
+			 *       up and doesn't find any more work it will
+			 *       again go back to waiting so no harm is
+			 *       done.
+			 */
+			kmbx = kse_wakeup_one(thread);
 		}
 	}
-        /*
-         * XXX - Threads are not yet assigned to specific KSEs; they are
-         *       assigned to the KSEG.  So the fact that a thread's KSE is
-         *       waiting doesn't necessarily mean that it will be the KSE
-         *       that runs the thread after the lock is granted.  But we
-         *       don't know if the other KSEs within the same KSEG are
-         *       also in a waiting state or not so we err on the side of
-         *       caution and wakeup the thread's last known KSE.  We
-         *       ensure that the threads KSE doesn't change while it's
-         *       scheduling lock is held so it is safe to reference it
-         *       (the KSE).  If the KSE wakes up and doesn't find any more
-         *       work it will again go back to waiting so no harm is done.
-         */
-	kse_wakeup_one(thread);
+	return (kmbx);
 }
 
-static void
+static struct kse_mailbox *
 kse_wakeup_one(struct pthread *thread)
 {
 	struct kse *ke;
@@ -1876,17 +1898,17 @@ kse_wakeup_one(struct pthread *thread)
 	if (KSE_IS_IDLE(thread->kse)) {
 		KSE_CLEAR_IDLE(thread->kse);
 		thread->kseg->kg_idle_kses--;
-		KSE_WAKEUP(thread->kse);
+		return (&thread->kse->k_mbx);
 	} else {
 		TAILQ_FOREACH(ke, &thread->kseg->kg_kseq, k_kgqe) {
 			if (KSE_IS_IDLE(ke)) {
 				KSE_CLEAR_IDLE(ke);
 				ke->k_kseg->kg_idle_kses--;
-				KSE_WAKEUP(ke);
-				return;
+				return (&ke->k_mbx);
 			}
 		}
 	}
+	return (NULL);
 }
 
 static void
diff --git a/lib/libpthread/thread/thr_mutex.c b/lib/libpthread/thread/thr_mutex.c
index c2c28e0..7e8de44 100644
--- a/lib/libpthread/thread/thr_mutex.c
+++ b/lib/libpthread/thread/thr_mutex.c
@@ -67,7 +67,8 @@
 /*
  * Prototypes
  */
-static void		mutex_handoff(struct pthread *, struct pthread_mutex *);
+static struct kse_mailbox *mutex_handoff(struct pthread *,
+			    struct pthread_mutex *);
 static inline int	mutex_self_trylock(struct pthread *, pthread_mutex_t);
 static inline int	mutex_self_lock(struct pthread *, pthread_mutex_t);
 static int		mutex_unlock_common(pthread_mutex_t *, int);
@@ -860,6 +861,7 @@ static int
 mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx = NULL;
 	int ret = 0;
 
 	if (m == NULL || *m == NULL)
@@ -904,7 +906,7 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 				 * Hand off the mutex to the next waiting
 				 * thread:
 				 */
-				mutex_handoff(curthread, *m);
+				kmbx = mutex_handoff(curthread, *m);
 			}
 			break;
 
@@ -961,7 +963,7 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 				 * Hand off the mutex to the next waiting
 				 * thread:
 				 */
-				mutex_handoff(curthread, *m);
+				kmbx = mutex_handoff(curthread, *m);
 			}
 			break;
 
@@ -1017,7 +1019,7 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 				 * Hand off the mutex to the next waiting
 				 * thread:
 				 */
-				mutex_handoff(curthread, *m);
+				kmbx = mutex_handoff(curthread, *m);
 			}
 			break;
 
@@ -1034,6 +1036,8 @@ mutex_unlock_common(pthread_mutex_t *m, int add_reference)
 
 		/* Unlock the mutex structure: */
 		THR_LOCK_RELEASE(curthread, &(*m)->m_lock);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 	}
 
 	/* Return the completion status: */
@@ -1460,9 +1464,10 @@ _mutex_lock_backout(struct pthread *curthread)
  * is necessary to lock the thread's scheduling queue while also
  * holding the mutex lock.
  */
-static void
+static struct kse_mailbox *
 mutex_handoff(struct pthread *curthread, struct pthread_mutex *mutex)
 {
+	struct kse_mailbox *kmbx = NULL;
 	struct pthread *pthread;
 
 	/* Keep dequeueing until we find a valid thread: */
@@ -1564,7 +1569,7 @@ mutex_handoff(struct pthread *curthread, struct pthread_mutex *mutex)
 		}
 
 		/* Make the thread runnable and unlock the scheduling queue: */
-		_thr_setrunnable_unlocked(pthread);
+		kmbx = _thr_setrunnable_unlocked(pthread);
 
 		/* Add a preemption point. */
 		if ((curthread->kseg == pthread->kseg) &&
@@ -1583,6 +1588,7 @@ mutex_handoff(struct pthread *curthread, struct pthread_mutex *mutex)
 	if ((pthread == NULL) && (mutex->m_protocol == PTHREAD_PRIO_INHERIT))
 		/* This mutex has no priority: */
 		mutex->m_prio = 0;
+	return (kmbx);
 }
 
 /*
diff --git a/lib/libpthread/thread/thr_private.h b/lib/libpthread/thread/thr_private.h
index 21026b7..90dbed7 100644
--- a/lib/libpthread/thread/thr_private.h
+++ b/lib/libpthread/thread/thr_private.h
@@ -1106,8 +1106,8 @@ void	_thr_ref_delete(struct pthread *, struct pthread *);
 int	_thr_schedule_add(struct pthread *, struct pthread *);
 void	_thr_schedule_remove(struct pthread *, struct pthread *);
 void	_thr_setrunnable(struct pthread *curthread, struct pthread *thread);
-void	_thr_setrunnable_unlocked(struct pthread *thread);
-void	_thr_sig_add(struct pthread *, int, siginfo_t *);
+struct kse_mailbox *_thr_setrunnable_unlocked(struct pthread *thread);
+struct kse_mailbox *_thr_sig_add(struct pthread *, int, siginfo_t *);
 void	_thr_sig_dispatch(struct kse *, int, siginfo_t *);
 int	_thr_stack_alloc(struct pthread_attr *);
 void	_thr_stack_free(struct pthread_attr *);
diff --git a/lib/libpthread/thread/thr_resume_np.c b/lib/libpthread/thread/thr_resume_np.c
index 1276311..ba9b911 100644
--- a/lib/libpthread/thread/thr_resume_np.c
+++ b/lib/libpthread/thread/thr_resume_np.c
@@ -35,7 +35,7 @@
 #include <pthread.h>
 #include "thr_private.h"
 
-static void resume_common(struct pthread *);
+static struct kse_mailbox *resume_common(struct pthread *);
 
 __weak_reference(_pthread_resume_np, pthread_resume_np);
 __weak_reference(_pthread_resume_all_np, pthread_resume_all_np);
@@ -46,15 +46,18 @@ int
 _pthread_resume_np(pthread_t thread)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx;
 	int ret;
 
 	/* Add a reference to the thread: */
 	if ((ret = _thr_ref_add(curthread, thread, /*include dead*/0)) == 0) {
 		/* Lock the threads scheduling queue: */
 		THR_SCHED_LOCK(curthread, thread);
-		resume_common(thread);
+		kmbx = resume_common(thread);
 		THR_SCHED_UNLOCK(curthread, thread);
 		_thr_ref_delete(curthread, thread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 	}
 	return (ret);
 }
@@ -64,6 +67,7 @@ _pthread_resume_all_np(void)
 {
 	struct pthread *curthread = _get_curthread();
 	struct pthread *thread;
+	struct kse_mailbox *kmbx;
 	kse_critical_t crit;
 
 	/* Take the thread list lock: */
@@ -73,8 +77,10 @@ _pthread_resume_all_np(void)
 	TAILQ_FOREACH(thread, &_thread_list, tle) {
 		if (thread != curthread) {
 			THR_SCHED_LOCK(curthread, thread);
-			resume_common(thread);
+			kmbx = resume_common(thread);
 			THR_SCHED_UNLOCK(curthread, thread);
+			if (kmbx != NULL)
+				kse_wakeup(kmbx);
 		}
 	}
 
@@ -83,7 +89,7 @@ _pthread_resume_all_np(void)
 	_kse_critical_leave(crit);
 }
 
-static void
+static struct kse_mailbox *
 resume_common(struct pthread *thread)
 {
 	/* Clear the suspend flag: */
@@ -95,5 +101,7 @@ resume_common(struct pthread *thread)
 	 * state to running and insert it into the run queue.
 	 */
 	if (thread->state == PS_SUSPENDED)
-		_thr_setrunnable_unlocked(thread);
+		return (_thr_setrunnable_unlocked(thread));
+	else
+		return (NULL);
 }
diff --git a/lib/libpthread/thread/thr_sig.c b/lib/libpthread/thread/thr_sig.c
index 1211ba3..fc075af 100644
--- a/lib/libpthread/thread/thr_sig.c
+++ b/lib/libpthread/thread/thr_sig.c
@@ -249,6 +249,7 @@ _thr_start_sig_daemon(void)
 void
 _thr_sig_dispatch(struct kse *curkse, int sig, siginfo_t *info)
 {
+	struct kse_mailbox *kmbx;
 	struct pthread *thread;
 
 	DBG_MSG(">>> _thr_sig_dispatch(%d)\n", sig);
@@ -280,9 +281,11 @@ _thr_sig_dispatch(struct kse *curkse, int sig, siginfo_t *info)
 			KSE_SCHED_UNLOCK(curkse, thread->kseg);
 			_thr_ref_delete(NULL, thread);
 		} else {
-			_thr_sig_add(thread, sig, info);
+			kmbx = _thr_sig_add(thread, sig, info);
 			KSE_SCHED_UNLOCK(curkse, thread->kseg);
 			_thr_ref_delete(NULL, thread);
+			if (kmbx != NULL)
+				kse_wakeup(kmbx);
 			break;
 		}
 	}
@@ -533,6 +536,7 @@ _thr_getprocsig_unlocked(int sig, siginfo_t *siginfo)
 struct pthread *
 thr_sig_find(struct kse *curkse, int sig, siginfo_t *info)
 {
+	struct kse_mailbox *kmbx = NULL;
 	struct pthread	*pthread;
 	struct pthread	*suspended_thread, *signaled_thread;
 	__siginfohandler_t *sigfunc;
@@ -582,7 +586,7 @@ thr_sig_find(struct kse *curkse, int sig, siginfo_t *info)
 				/*  where to put siginfo ? */
 				*(pthread->data.sigwaitinfo) = si;
 				pthread->sigmask = pthread->oldsigmask;
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 			}
 			KSE_SCHED_UNLOCK(curkse, pthread->kseg);
 			/*
@@ -596,6 +600,8 @@ thr_sig_find(struct kse *curkse, int sig, siginfo_t *info)
 			 * to the process pending set.
 			 */
 			KSE_LOCK_RELEASE(curkse, &_thread_list_lock);
+			if (kmbx != NULL)
+				kse_wakeup(kmbx);
 			return (NULL);
 		} else if (!SIGISMEMBER(pthread->sigmask, sig) ||
 			   (!SIGISMEMBER(pthread->oldsigmask, sig) &&
@@ -811,15 +817,16 @@ handle_special_signals(struct kse *curkse, int sig)
  *
  * This must be called with the thread's scheduling lock held.
  */
-void
+struct kse_mailbox *
 _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 {
+	siginfo_t siginfo;
+	struct kse *curkse;
+	struct kse_mailbox *kmbx = NULL;
+	struct pthread *curthread = _get_curthread();
 	int	restart;
 	int	suppress_handler = 0;
 	int	fromproc = 0;
-	struct  pthread *curthread = _get_curthread();
-	struct	kse *curkse;
-	siginfo_t siginfo;
 
 	DBG_MSG(">>> _thr_sig_add %p (%d)\n", pthread, sig);
 
@@ -829,12 +836,12 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 
 	if (pthread->state == PS_DEAD || pthread->state == PS_DEADLOCK ||
 	    pthread->state == PS_STATE_MAX)
-	    	return; /* return false */
+	    	return (NULL); /* return false */
 
 	if ((pthread->attr.flags & PTHREAD_SCOPE_SYSTEM) &&
 	    (curthread != pthread)) {
 	    	PANIC("Please use _thr_send_sig for bound thread");
-		return;
+		return (NULL);
 	}
 
 	if (pthread->curframe == NULL ||
@@ -851,7 +858,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 					 sizeof(*info));
 		} else {
 			if (!_thr_getprocsig(sig, &pthread->siginfo[sig-1]))
-				return;
+				return (NULL);
 			SIGADDSET(pthread->sigpend, sig);
 		}
 		if (!SIGISMEMBER(pthread->sigmask, sig)) {
@@ -867,7 +874,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 		/* if process signal not exists, just return */
 		if (fromproc) {
 			if (!_thr_getprocsig(sig, &siginfo))
-				return;
+				return (NULL);
 			info = &siginfo;
 		}
 		/*
@@ -877,7 +884,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 		case PS_DEAD:
 		case PS_DEADLOCK:
 		case PS_STATE_MAX:
-			return;	/* XXX return false */
+			return (NULL);	/* XXX return false */
 		case PS_LOCKWAIT:
 		case PS_SUSPENDED:
 			/*
@@ -937,7 +944,7 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 				*(pthread->data.sigwaitinfo) = pthread->siginfo[sig-1];
 				pthread->sigmask = pthread->oldsigmask;
 				/* Make the thread runnable: */
-				_thr_setrunnable_unlocked(pthread);
+				kmbx = _thr_setrunnable_unlocked(pthread);
 			} else {
 				/* Increment the pending signal count. */
 				SIGADDSET(pthread->sigpend, sig);
@@ -945,11 +952,10 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 					pthread->check_pending = 1;
 					pthread->interrupted = 1;
 					pthread->sigmask = pthread->oldsigmask;
-					_thr_setrunnable_unlocked(pthread);
+					kmbx = _thr_setrunnable_unlocked(pthread);
 				}
 			}
-		
-			return;
+			return (kmbx);
 		}
 
 		SIGADDSET(pthread->sigpend, sig);
@@ -967,11 +973,12 @@ _thr_sig_add(struct pthread *pthread, int sig, siginfo_t *info)
 			if (pthread->flags & THR_FLAGS_IN_RUNQ)
 				THR_RUNQ_REMOVE(pthread);
 			pthread->active_priority |= THR_SIGNAL_PRIORITY;
-			_thr_setrunnable_unlocked(pthread);
+			kmbx = _thr_setrunnable_unlocked(pthread);
 		} else {
 			pthread->check_pending = 1;
 		}
 	}
+	return (kmbx);
 }
 
 /*
@@ -981,6 +988,7 @@ void
 _thr_sig_send(struct pthread *pthread, int sig)
 {
 	struct pthread *curthread = _get_curthread();
+	struct kse_mailbox *kmbx;
 
 	if (pthread->attr.flags & PTHREAD_SCOPE_SYSTEM) {
 		kse_thr_interrupt(&pthread->tmbx, KSE_INTR_SENDSIG, sig);
@@ -990,8 +998,10 @@ _thr_sig_send(struct pthread *pthread, int sig)
 	/* Lock the scheduling queue of the target thread. */
 	THR_SCHED_LOCK(curthread, pthread);
 	if (_thread_sigact[sig - 1].sa_handler != SIG_IGN) {
-		_thr_sig_add(pthread, sig, NULL);
+		kmbx = _thr_sig_add(pthread, sig, NULL);
 		THR_SCHED_UNLOCK(curthread, pthread);
+		if (kmbx != NULL)
+			kse_wakeup(kmbx);
 		/* XXX
 		 * If thread sent signal to itself, check signals now.
 		 * It is not really needed, _kse_critical_leave should